2012 International Conference for High Performance Computing, Networking, Storage and Analysis · 2012 · 33 citations · 22 references
ComputingComputer ScienceHigh-performance ComputingParallel AlgorithmsParallel ComputingScientific ComputingData ParallelismGraphics Processing Unit ComputingGraphics Processing UnitsMatrix ComputationNumerical Linear AlgebraParallel Numerical MethodsDivergenceGpu Parallel AlgorithmsMemory Footprint ReductionParallel Matrix ComputationParallel SolversScientific Computing On GpusSparse Matrix ComputationTiling Strategies
In this paper, we present a scalable, numerically stable, high-performance tridiagonal solver. The solver is based on the SPIKE algorithm for partitioning a large matrix into small independent matrices, which can be solved in parallel. For each small matrix, our solver applies a general 1-by-1 or 2-by-2 diagonal pivoting algorithm, which is also known to be numerically stable. Our paper makes two major contributions. First, our solver is the first numerically stable tridiagonal solver for GPUs. Our solver provides comparable quality of stable solutions to Intel MKL and Matlab, at speed comparable to the GPU tridiagonal solvers in existing packages like CUSPARSE. It is also scalable to multiple GPUs and CPUs. Second, we present and analyze two key optimization strategies for our solver: a high-throughput data layout transformation for memory efficiency, and a dynamic tiling approach for reducing the memory access footprint caused by branch divergence.
22
Elementary Numerical Analysis: An Algorithmic Approach
J. W. W., S. D. Conte · Mathematics of Computation · 1967 · 1.3K citations
A Fast Direct Solution of Poisson's Equation Using Fourier Analysis
R. W. Hockney · Journal of the ACM · 1965 · 692 citations · Full text
Scan Primitives for GPU Computing
eScholarship (California Digital Library) · 2007 · 513 citations · Full text